This analysis investigates the relationship between food access and health outcomes across North Carolina census tracts, using USDA Food Access Research Atlas and CDC PLACES health data. The study explores how living in food deserts correlates with obesity and diabetes prevalence, while accounting for socioeconomic factors. For a map of food deserts and food swamps in North Carolina, please view my supplemental map here (takes some time to load): https://johnetobin.shinyapps.io/foodInsecurityMap/
Key findings include:
This evidence suggests that addressing food access should be a key component of public health initiatives in North Carolina, particularly when implemented alongside poverty reduction strategies.
The violin plots reveal the distribution of both obesity and diabetes rates by food desert status. The plots show not only higher mean values in food desert areas but also different distribution patterns. Food desert areas display wider variability in health outcomes, particularly in the upper ranges, suggesting more extreme cases in these communities. Statistical analysis confirms these visual patterns:
Obesity: Food desert tracts show significantly higher obesity rates (t-test: p < 0.001) Diabetes: The pattern holds for diabetes as well (t-test: p < 0.001)
These plots show that food desert status, poverty rate, and median family income are all associated with obesity rates, suggesting a complex relationship between socioeconomic factors, food access, and health outcomes.
This boxplot reveals a crucial insight: the relationship between food desert status and obesity varies across different poverty levels. The gap in obesity rates between food desert and non-food desert areas is particularly pronounced in areas with moderate poverty (10-20%), while narrowing slightly in the highest poverty areas. This suggests that:
My analysis identified several key dimensions of food access that explain 88% of the variance in the food access variables:
These biplots provide several key
insights:
The clustering of high obesity areas (dark red points) primarily in the upper right quadrant of the first plot indicates that combinations of access dimensions (rather than single factors) are associated with the worst health outcomes.
Food desert tracts cluster differently from non-food desert tracts, confirming that the PCA has effectively captured meaningful patterns in food access.
The spread of both food desert and non-food desert points across different regions suggests that even within these categories, there is considerable variability in access patterns that may be relevant for intervention design.
| Model | R_squared | Adj_R_squared | F_statistic | p_value |
|---|---|---|---|---|
| Model 1: Food Desert Only | 0.140 | 0.139 | 124.28 | < 2.2e-16 |
| Model 2: + Socioeconomic | 0.630 | 0.628 | 426.85 | < 2.2e-16 |
| Model 3: PCA Components | 0.471 | 0.467 | 134.80 | < 2.2e-16 |
| Model 4: Full Model | 0.650 | 0.647 | 173.97 | < 2.2e-16 |
| Model | R_squared | Adj_R_squared | F_statistic | p_value |
|---|---|---|---|---|
| Model 1: Food Desert Only | 0.085 | 0.084 | 71.16 | < 2.2e-16 |
| Model 2: + Socioeconomic | 0.381 | 0.378 | 154.20 | < 2.2e-16 |
| Model 3: PCA Components | 0.360 | 0.355 | 85.13 | < 2.2e-16 |
| Model 4: Full Model | 0.468 | 0.462 | 82.13 | < 2.2e-16 |
My regression analysis progressed through increasingly complex models:
Model 1: Only food desert status (R² = 0.14 for obesity, 0.09 for diabetes)
Model 2: Food desert status + socioeconomic factors (R² = 0.63 for obesity, 0.38 for diabetes)
Model 3: PCA components only (R² = 0.47 for obesity, 0.36 for diabetes)
Model 4: Full model with all predictors (R² = 0.65 for obesity, 0.47 for diabetes)
The substantial improvement in model fit from Model 1 to Model 2 highlights the importance of socioeconomic context, while the strong performance of Model 3 demonstrates that food access dimensions capture significant health variation independent of the binary food desert classification.
Based on these findings, effective public health strategies should:
This analysis has several limitations that provide opportunities for future research:
1: Causality: The cross-sectional design limits causal inference about the relationship between food access and health outcomes. 2. Urban Focus: Data availability constraints limited the analysis primarily to urban census tracts, necessitating future dedicated studies of rural food environments. 3. Individual Behaviors: Area-based measures don’t capture individual food shopping behaviors, dietary choices, or transportation patterns.
Future research should explore:
This analysis provides compelling evidence that food access is an important social determinant of health in North Carolina communities. The multidimensional nature of food access revealed through PCA suggests that solutions must be similarly multifaceted, addressing not just store proximity but the full range of barriers that limit access to healthy food options. By targeting these access dimensions alongside socioeconomic factors, public health initiatives may more effectively reduce obesity and diabetes disparities across the state.